Added support DLL (dynamic link library) - yolo_cpp_dll.dll

AlexeyAB

2017-03-15 a6cbaeecde40f91ddc3ea09aa26a03ab5bbf8ba8

 README.md

@@ -5,9 +5,13 @@
3. [How to train (Pascal VOC Data)](#how-to-train-pascal-voc-data)
4. [How to train (to detect your custom objects)](#how-to-train-to-detect-your-custom-objects)
5. [When should I stop training](#when-should-i-stop-training)
6. [How to mark bounded boxes of objects and create annotation files](#how-to-mark-bounded-boxes-of-objects-and-create-annotation-files)
6. [How to improve object detection](#how-to-improve-object-detection)
7. [How to mark bounded boxes of objects and create annotation files](#how-to-mark-bounded-boxes-of-objects-and-create-annotation-files)

|  ![Darknet Logo](http://pjreddie.com/media/files/darknet-black-small.png) |   ![map_fps](https://cloud.githubusercontent.com/assets/4096485/21550284/88f81b8a-ce09-11e6-9516-8c3dd35dfaa7.jpg) https://arxiv.org/abs/1612.08242 |
|  ![Darknet Logo](http://pjreddie.com/media/files/darknet-black-small.png) |   ![map_fps](https://hsto.org/files/a24/21e/068/a2421e0689fb43f08584de9d44c2215f.jpg) https://arxiv.org/abs/1612.08242 |
|---|---|

|  ![Darknet Logo](http://pjreddie.com/media/files/darknet-black-small.png) |   ![map_fps](https://hsto.org/files/978/a64/7ca/978a647caaee40b7b0a64f7770f11e99.jpg) https://arxiv.org/abs/1612.08242 |
|---|---|


@@ -291,6 +295,18 @@
| ![Yolo_v2_training](https://hsto.org/files/d12/1e7/515/d121e7515f6a4eb694913f10de5f2b61.jpg) | ![Yolo_v2_training](https://hsto.org/files/727/c7e/5e9/727c7e5e99bf4d4aa34027bb6a5e4bab.jpg) |
|---|---|

## How to improve object detection:

1. Before training:
  * set flag `random=1` in your `.cfg`-file - it will increase precision by training Yolo for different resolutions: [link](https://github.com/AlexeyAB/darknet/blob/47409529d0eb935fa7bafbe2b3484431117269f5/cfg/yolo-voc.cfg#L244)

2. After training - for detection:

  * Increase network-resolution by set in your `.cfg`-file (`height=608` and `width=608`) or (`height=832` and `width=832`) or (any value multiple of 32) - this increases the precision and makes it possible to detect small objects: [link](https://github.com/AlexeyAB/darknet/blob/47409529d0eb935fa7bafbe2b3484431117269f5/cfg/yolo-voc.cfg#L4)
  
    * you do not need to train the network again, just use `.weights`-file already trained for 416x416 resolution
    * if error `Out of memory` occurs then in `.cfg`-file you should increase `subdivisions=16`, 32 or 64: [link](https://github.com/AlexeyAB/darknet/blob/47409529d0eb935fa7bafbe2b3484431117269f5/cfg/yolo-voc.cfg#L3)

## How to mark bounded boxes of objects and create annotation files:

Here you can find repository with GUI-software for marking bounded boxes of objects and generating annotation files for Yolo v2: https://github.com/AlexeyAB/Yolo_mark